Introduction

Post-colonoscopy colorectal cancers (PCCRCs) represent missed opportunities to diagnose or prevent colorectal cancer. Up to 89% of PCCRCs are avoidable [1], but identifying PCCRCs for root-cause analysis can be difficult, especially when the cancer is diagnosed in a different hospital from the index colonoscopy.

Aims & Methods

The aim of this study was to create a process for identifying PCCRCs for all English National Health Service (NHS) hospitals, introduce a case review template to identify the most plausible explanation for PCCRC, and use the evidence captured to inform recommendations for reducing PCCRC rates.
PCCRCs occurring 6-48 months after an index colonoscopy (PCCRC-4yr) were identified by linking datasets (cancer registry and hospital episode statistics) and uploaded into a secure portal with an audit template based on World Endoscopy Organisation (WEO) recommendations and a previously published case review [1]. WEO categories were applied to determine a plausible aetiology based on whether adenomas were detected in the same segment (adequate examination (A), inadequate examination (B)), and if a lesion was resected (no (C), yes, and likely incomplete resection (D)). PCCRC were also stratified based on whether they occurred before (interval), at (non-interval A), after a planned surveillance procedure (non-interval B), or where no surveillance was planned (non-interval C). This method was tested at 10 pilot sites, refined, and then rolled out nationally. A maximum of 25 PCCRCs were uploaded to 126 hospital sites.

Results

2,465 PCCRCs were uploaded to the portal. 122/126 sites participated. In total, 459 patients were excluded from the study after the initial review. A complete audit dataset was available for 1724 of the 2006 remaining PCCRCs (86% compliance). In 294 of the 1724 (17%) audited PCCRCs, index colonoscopy was performed in a different organisation from where the cancer was diagnosed. Overall, 1190 (70%) were potentially avoidable (4% definitely, 13% probably, and 53% possibly). 16% of PCCRCs were judged to have caused major harm or premature death due to a delay in diagnosis. Photographic evidence of completion and rectal retroflexion was absent or inadequate in >50% of cases, where it was deemed important. There were 1179 WEO A (68%), 308 WEO B (18%), 157 WEO C (9%), and 80 WEO B (5%). There were 315 interval PCCRC (18%), 280 non-interval A PCCRCs (16%), 179 non-interval B PCCRCs (10%), 773 non-interval C PCCRCs (45%), and 177 unknown PCCRCs (10%). Patient, clinical, and administrative factors were involved in 10 %, 10%, and 5% of cases, respectively.

Conclusion

A process for identifying PCCRCs in the English NHS has shown that 17% of cases are diagnosed in other hospitals, cases that would not normally be known to the site of the index colonoscopy. This initial analysis of completed audits of 1724 PCCRCs provides the largest dataset of aggregated data on PCCRCs and the only one for an entire country. This confirms that the majority of PCCRCs are not just avoidable, but the delay leads to moderate-to-severe harm and even premature death in a large proportion of patients. Unlike multivariate analyses of large datasets, it provides detailed information on what leads to delays in diagnosis and what needs to be done to reduce the number of PCCRCs. Large dataset analyses are important to highlight potential variations in colonoscopy performance, but detailed case reviews are needed to determine the aetiology of PCCRC, provide a mechanism to provide feedback to colonoscopists and patients, and formulate improvement plans.

References

Anderson, R, Burr, N, Valori, R. Gastroenterology 2020;158:1287-1299

Introduction

Anemia is a common and often underdiagnosed condition in the elderly population. Digestive endoscopy plays a pivotal role in the etiological workup, particularly to rule out serious underlying causes such as malignancies or peptic lesions. However, the diagnostic yield of such procedures in routine clinical practice remains variable and underreported, especially in resource-limited settings.

Aims & Methods

To assess the diagnostic yield of digestive endoscopy in the investigation of anemia in elderly patients, with a particular focus on iron deficiency anemia (IDA) and vitamin B12 deficiency, and to identify specific endoscopic findings associated with these etiologies.
We conducted a retrospective descriptive study over a 6-month period. The study included all patients aged 65 years and older who underwent esophagogastroduodenoscopy (EGD) and/or colonoscopy as part of the diagnostic evaluation for anemia. Data collected included demographic information, clinical symptoms, laboratory parameters (including hemoglobin, ferritin, vitamin B12 levels), endoscopic findings, and histopathological results when available.

Results

A total of 252 patients were included in the analysis. Among them, 44 patients (17.5%) with a mean age of 71 years [68-82], had confirmed IDA, and six patients (2.4%) presented with vitamin B12 deficiency anemia, associated in all cases with histologically confirmed fundic atrophy.
In univariate analysis, IDA was statistically correlated with the presence of peptic esophagitis (p=0.04), gastric neoplasia (p=0.001), erosive gastropathy (p=0.02), and Helicobacter pylori gastritis (p=0.04). It was also weakly correlated with the presence of colorectal cancer (p=0.05).
Colonic angiodysplasias and colonic polyps were not associated with IDA in this cohort (p=0.1 and p=0.4, respectively).

Conclusion

Digestive endoscopy, particularly EGD, remains a cornerstone in the diagnostic algorithm of anemia in the elderly. It enables the identification of a wide range of underlying pathologies, from inflammatory lesions to malignancies, and guides appropriate therapeutic interventions. Early endoscopic evaluation should be encouraged in this population to prevent diagnostic delays and improve outcomes.

Introduction

Serrated lesions in the colon, that can contribute to up to 20% of colon cancers, are often difficult to detect during standard white light colonoscopy (SC). Artificial intelligence specifically Computer-aided Detection (CADe) has been shown to positively impact the adenoma detection rate (ADR) and adenomas per colonoscopy, but data on SSLs are scarce. We aimed to evaluate the effect of CADe on detecting SSLs byperforming a meta-analysis of all the randomized controlled trials.

Aims & Methods

PubMed, Embase, and Cochrane databases were searched (upto November 2022) to identify randomized controlled studies comparing CADe and SC. Our primary outcome was the comparison of pooled estimates of SSL-detection rate (SSLDR) between those undergoing CADe and SC. The secondary outcome was the mean number of SSL per colonoscopy (SSLPC). Subgroup analyses were performed for studies only including screening or surveillance colonoscopy. Risk ratios (RR) and mean differences (MD) with 95% confidence intervals (CI) were calculated for dichotomous and continuous variables, respectively, using the random effects model. Study heterogeneity was assessed using i² statistics and prediction intervals

Results

12 RCTs were eligible with a total of 9237 patients: CADe group-4595 patients (mean age 58.5 years; 53.5% men) and SC group- 4642 (mean age 58.4 years; 52.1% men). There was no significant difference in SSLDR between CADe (4.6%) vs. SC (4%) with a Risk Ratio of 1.15 (95% CI 0.95-1.39, i²= 0%) (Table 1a). Similarly, for SSLPC, there was no significant difference between CADe and SC (MD 0.01, 95% CI -0.02 –0.03, i²= 98%). When our analysis was limited to studies including only screening or surveillance colonoscopy, again, there were no significant differences between CADe and SC in terms of SSLDR (5.7% vs 4.9%, RR 1.15, 95% CI 0.92-1.45) and SSLPC (MD -0.02, 95% CI -0.07 – 0.030).

Table 1. Comparison of SSLDR between CADe and Standard Colonoscopy

Outcome		No of Patients		SSLDR (%)		RR	95% CI	Prediction Interval	I2 %
		CADe	SC	CADe	SC
SSLDR	All Studies	4595	4642	4.6	4.0	1.15	0.95 – 1.39	0.93 - 1.42	0
	Screening/Surveillance Only	2475	2524	5.7	4.9	1.15	0.92 - 1.45	0.80 - 1.67	0

Conclusion

Compared with standard colonoscopy, currently evaluated AI systems do not show improved sessile serrated lesion detection rate or mean sessile serrated lesions per colonoscopy, even when including only screening or surveillance colonoscopy. Further RCTs should be designed with SSL as one of the primary outcomes using highly trained AI systems.

Disclosure

No Relevant Financial Disclosure

Introduction

Hepatitis C virus (HCV) infection continues to pose a significant public health challenge, with demographic trends shaped by migration patterns and shifting risk profiles. Gaining insight into these dynamics is essential for optimising treatment approaches and informing effective resource allocation.

Aims & Methods

This retrospective cohort study aimed to analyze demographic patterns, HCV genotype distribution, and cirrhosis prevalence among patients treated for hepatitis C at St. Luke’s Hospital, Ireland, between 2016 and 2024. Statistical analyses included descriptive metrics, Chi-Square tests, one-way ANOVA, and multinomial logistic regression, performed using SPSS.

Results

A total of 202 patients received treatment during the study period, with 67% identifying as male and a median age of 46 years. The majority were born in Ireland (45.5%), Eastern Europe (16.8%), or other European countries (15.3%). No significant differences were observed in age at presentation across years (p = 0.449) or by country of birth (p = 0.367). Gender distribution remained stable over time (p = 0.373) and across birth countries (p = 0.107). Genotype distribution, however, varied significantly by country of birth (p < 0.001): Genotype 1b was more prevalent among patients from Eastern and other European countries, while Genotypes 1a and 3 were more common among Irish-born individuals. Genotypes 3 and 4 were seen more frequently in patients from non-European regions. Cirrhosis prevalence showed no significant association with genotype (p = 0.071) or country of birth (p = 0.064). Although country of origin did not change significantly over time (p = 0.224), descriptive trends indicated an increase in non-Irish-born patients.

Conclusion

This study highlights consistent demographic trends in HCV presentations over time, alongside distinct regional variations in genotype distribution. Continued monitoring of these patterns is essential for guiding service planning and delivering targeted, patient-centered care in a changing population landscape.

Introduction

Optical diagnosis of colorectal polyps is challenging. Artificial intelligence based systems allow the detection and histological characterization of polyps.

Aims & Methods

The aim of this study is to compare the diagnostic accuracy of Polydeep, a CADe/x system with expert endoscopists for the optical diagnosis of colorectal polyps.
We designed a diagnostic test study based on the evaluation of 2364 polyps still images with confirmed histological diagnosis (69.12% adenomas, 14.8% Traditional Serrated Adenoma, TSA, or Serrated Sessile Lesion, SSL, 16.7% hyperplastic). The lesions was classificated as neoplastic (adenomatous, TSA, SSL) and no neoplastic (hyperplastic lesions). The images were shown to Polydeep and to the expert endoscopists to obtain the predicted histology. The endoscopists classified the lesions as adenoma, TSA, SSL or hyperplastic lesions, while PolyDeep classified the lesions as neoplastic and non-neoplastic. We calculated sensitivity, specificity, predictive values, likelihood ratios, diagnostic odds ratio and the Area Under the Curve (AUC). We compared sensitivity and specificity using the McNemar test and the AUC using the homogeneity areas test.

Results

PolyDeep classified 77.62% of evaluated lesions as neoplastic and the endoscopists classified 87.1% of the lesions as neoplastic (adenomas: 75.76%, TSA: 3.86%, SSL: 7.55%). The diagnostic accuracy results as well as the comparison between PolyDeep and the endoscopists for the neoplastic category is shown in the attached table.

	Endoscopists	PolyDeep	P-value
Sensitivity	91.56%	90.34%	<0.001
Specificity	35.23%	35.31%	0.355
PPV	88.21%	88.72%
NPV	44.07%	39.37%
LR +	1.41	1.39
LR -	0.24	0.27
OR	5.89	5.11
AUC	0.634	0.628	0.91

Table 1: The table show the comparison of diagnostic test between endoscopists and PolyDeep, furthermore the AUC and p-values of McNemar test and the comparison between both ROC curves. Positive Predictive Value (PPV), Negative predictive Value (NPV), positive Likelihood Ratio (LR+), negative Likelihood Ratio (LR-), Odds Ratio (OR), Area Under the Curve (AUC).

Conclusion

Polydeep, a CADe/x system, is as accurate as expert endoscopists for the optical diagnosis of colorectal polyps.

Introduction

Nowadays there is no screening test approved on large scale for gastric cancer even if this neoplasia represents one of the leading causes of cancer-related deaths globally^1-2. In Italy still remains one of the most frequent tumors, with incidence rates differing across geographical areas³.

Aims & Methods

This project aims to create an easily administrable tool to detect anamnestic gastric cancer risk factors together with alarm symptoms in an age-defined target population to identify patients in need of an esophagogastroduodenoscopy (EGD).
This paper describes the first study phase: the administration of the questionnaire to a first sample of 5000 people aged from 40 to 80 years and the preliminary results. GASTROFORM is an online questionnaire (https://it.research.net/r/GASTROFORM) collecting data about demographic and anthropometric features, physiological anamnesis, familial history, non-specific upper gastrointestinal symptomatology, alarm symptoms and recent EGD performance.

Results

4426 valid questionnaires were submitted through the online form in November 2022, 612 of them (13,82%) reported at least one alarm symptom identifying patients in need to undergo an EGD as soon as possible. From an epidemiological point of view we expect to increase the rate of diagnosis in Barret esophagus (from 1,8% to 5-8%), atrophic gastritis (from 9% to 20-30%) and early gastric cancer (from 0,019% to 1-2 %).

Conclusion

The GASTROFORM questionnaire could provide useful informations to identify a population most likely to have significant endoscopic findings in early diagnosis during a screening EGD.

References

1. Ilic M, Ilic I. Epidemiology of stomach cancer. World J Gastroenterol. 2022;28(12):1187-1203. doi:10.3748/wjg.v28.i12.1187
2. Rawla P, Barsouk A. Epidemiology of gastric cancer: global trends, risk factors and prevention. Prz Gastroenterol. 2019;14(1):26-38. doi:10.5114/pg.2018.80001
3. Donida BM, Tomasello G, Ghidini M, et al. Epidemiological, clinical and pathological characteristics of gastric neoplasms in the province of Cremona: the experience of the first population-based specialized gastric cancer registry in Italy. BMC Cancer. 2019;19(1):212. Published 2019 Mar 8. doi:10.1186/s12885-019-5366-1

Introduction

The advancement of computer-aided detection systems using artificial intelligence (AI) opened a new scope in endoscopic image analysis. However, the deep-learning-based models developed on closed datasets are susceptible to unknown anomalies in complex clinical environments. In particular, the high false positive rate in polyp detection remains a major challenge in clinical practice. The frequent occurrence of false positives (FPs) leads to endoscopists' fatigue, distraction, and the need for refocusing. It costs additional time and effort to discriminate FPs from true positives (TPs).

Aims & Methods

We propose a generic and practical solution for FP suppression in real-world AI-assisted colonoscopy. We release the FPPD-13 dataset, which includes real-world cases of TPs (1300 samples) and 13 classes of FPs (100 samples/class) with a comprehensive taxonomy. A YOLOv5 polyp detector was trained on a private dataset collected in Zhongshan Hospital. It was employed to analyze real-world colonoscopy videos for the collection of TPs and FPs. Two endoscopists discriminated all the positive predictions into TPs and FPs, and a third senior endoscopist was consulted for controversial images. We further propose EndoBoost, a plug-and-play module for FP suppression during polyp detection. EndoBoost follows the formulation of anomaly detection (AD) and takes FPs as anomalies. Specifically, a normalizing flow is utilized for density estimation in the feature space and rejecting FPs during real-time inference. We develop a learnable image encoder to obtain an informative feature space for the AD task. The image encoder and the normalizing flow are jointly optimized to learn the TP manifold while the FP samples are also exploited through outlier exposure.

Results

	AP	AUC	Accuracy	Precision	Sensitivity	Specificity
EndoBoost	0.788±0.017	0.793±0.020	0.705±0.019	0.655±0.026	0.877±0.061	0.534±0.078
KDE	0.646±0.024	0.610±0.027	0.569±0.021	0.541±0.016	0.928±0.036	0.209±0.076
OC-SVM	0.733±0.011	0.726±0.004	0.627±0.037	0.588±0.036	0.886±0.071	0.368±0.145
MCD	0.700±0.009	0.730±0.004	0.657±0.009	0.604±0.011	0.915±0.036	0.398±0.051
PCA (10)	0.654±0.020	0.677±0.013	0.611±0.017	0.567±0.012	0.948±0.024	0.275±0.047
PCA (100)	0.690±0.013	0.721±0.005	0.657±0.008	0.605±0.008	0.902±0.030	0.412±0.035
AE (128)	0.647±0.031	0.675±0.025	0.605±0.022	0.561±0.014	0.960±0.013	0.249±0.044
VAE (128)	0.718±0.008	0.703±0.008	0.588±0.028	0.557±0.025	0.902±0.062	0.275±0.118
iForest	0.644±0.034	0.634±0.026	0.577±0.015	0.544±0.010	0.953±0.025	0.200±0.046

Extensive experiments demonstrate a promising false positive suppression in both retrospective and prospective validation. EndoBoost was validated and compared to other competitors on the proposed dataset. We first evaluated the performance of different AD models when only TPs were accessible during training. EndoBoost achieved the highest AP and AUC among all the AD models (Table). Data-efficiency experiments were conducted to explore how to make the most use of available FPs during training. EndoBoost reached the equivalent performance (AP: 0.965, AUC: 0.966) with the pervious fully supervised method using only 10% FPs. For robustness to unseen classes of false positives, EndoBoost outperformed others with the highest AP (0.781) and AUC (0.974). In the prospective study, we took three colonoscopic video clips with a total length of about 1 hour as the real-world deployment test set. With the assistance of EndoBoost, FPs were suppressed successfully during the withdrawal of the colonoscopy.

Conclusion

In this work, we present a practical solution of dataset and methodology for reducing FPs during AI-assisted colonoscopy. The FPPD-13 dataset and proposed EndoBoost demonstrate their effectiveness in supressing FPs during real-world colonoscopy. For future work, multi-center clinical trials using EndoBoost are being carried out.